AMR_lic/amr_example.py at master · silvianacmp/AMR_lic · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
from __future__ import print_function
import dynet as dy
import deep_dynet.support as ds
import deep_dynet.transition_parser as ddtp
import logging
from smatch import smatch_amr
from smatch import smatch_util
import numpy as np
import matplotlib.pyplot as plt
import traceback
from amr_util.Reporting import AMRResult
import amr_util.Actions as act
from amr_reader import read_data


def process_data(data, vocab_words, vocab_acts):
    for d in data:
        sentence = d[0]
        actions = d[1]
        amr_str = d[2]
        concept_meta = d.concepts_metadata
        yield (
            ds.word_sentence_to_vocab_index(sentence.split(), vocab_words),
            actions,
            sentence,
            actions,
            amr_str,
            concept_meta
        )

logging.basicConfig(format='%(asctime)s %(levelname)s: %(message)s', level=logging.WARNING)

vocab_acts = ds.Vocab.from_list(act.acts)
vocab_words = ds.Vocab.from_file('resources/data/vocab.txt')

test_use_model_prediction = True
train_use_model_prediction = False
tests = ["dfa"]
# tests = ["bolt", "dfa", "proxy", "xinhua", "deft"]
# tests = ["deft"]
cases = []
for filter_path in tests:
    training_data = read_data("training",cache=False, filter_path=filter_path)
    # dev_data = read_data("dev")
    test_data = read_data("dev", cache = False, filter_path=filter_path)
    print("%s Training size %d" % (filter_path, len(training_data)))
    print("%s Test size %d" % (filter_path, len(test_data)))


    # train = list(ds.read_oracle('resources/data/amr-examples.txt', vocab_words, vocab_acts))
    # dev = list(ds.read_oracle('resources/data/amr-examples-test.txt', vocab_words, vocab_acts))
    train = list(process_data(training_data, vocab_words, vocab_acts))
    test = list(process_data(test_data, vocab_words, vocab_acts))
    cases.append((filter_path, train, test))

amr_dynet_results = []
for run in range(1):
    for (filter_path, train, test) in cases:
        model = dy.Model()
        trainer = dy.AdamTrainer(model)

        tp = ddtp.TransitionParser(model, vocab_words)
        logging.info("Processing %s at run %s", filter_path, run)

        # cmake .. -DEIGEN3_INCLUDE_DIR=/Users/flo/Documents/Doctorat/AMR/dynet-base/eigen -DBOOST_ROOT=/usr/local/opt/boost160/ -DPYTHON=/usr/bin/python
        accuracies = []
        rounds = 0
        best_epoch = 0
        fail_sentences = []
        right_predictions = 0.0
        total_predictions = 0
        for epoch in range(10):
            smatch_train_results = smatch_util.SmatchAccumulator()
            for (sentence, actions, original_sentence, original_actions, amr, concepts_metadata) in train:
                loss = None
                try:
                    parsed = tp.parse(sentence, actions, concepts_metadata, use_model_predictions=train_use_model_prediction)
                    loss = parsed[0]
                    parsed_amr = parsed[1]
                    right_predictions += parsed[2]
                    total_predictions += parsed[3]

                    parsed_amr_str = parsed_amr.amr_print()
                    original_amr = smatch_amr.AMR.parse_AMR_line(amr)
                    parsed_amr = smatch_amr.AMR.parse_AMR_line(parsed_amr_str)
                    smatch_f_score = smatch_train_results.compute_and_add(parsed_amr, original_amr)

                except Exception as e:
                    logging.warn(e)
                    fail_sentences.append(original_sentence)
                    logging.warn("%s\n with actions %s\n", original_sentence, original_actions)
                    traceback.print_exc()
                if loss is not None:
                    # for some weird reason backward throws an failed assertion if there is no scalar value retrievall
                    loss.scalar_value()
                    loss.backward()
                    trainer.update()

            accuracy = 0
            if total_predictions > 0:
                accuracy = right_predictions / total_predictions

            hist, bins = np.histogram(smatch_train_results.smatch_scores, bins=AMRResult.histogram_beans())
            train_result = AMRResult(filter_path, epoch, "train", len(train), accuracy, -1, bins, hist, smatch_train_results.get_result())
            dev_words = 0
            dev_loss = 0.0
            right_predictions = 0.0
            total_predictions = 0
            invalid_actions = 0
            fail_sentences = []

            smatch_test_results = smatch_util.SmatchAccumulator()
            for (ds, da, original_sentence, original_actions, amr, concepts_metadata) in test:
                loss = None
                try:
                    parsed_sentence = tp.parse(ds, da, concepts_metadata, use_model_predictions=test_use_model_prediction)
                    loss = parsed_sentence[0]
                    parsed_amr = parsed_sentence[1]
                    right_predictions += parsed_sentence[2]
                    total_predictions += parsed_sentence[3]
                    invalid_action = parsed_sentence[5]
                    invalid_actions += invalid_action
                    dev_words += len(ds)

                    parsed_amr_str = parsed_amr.amr_print()
                    original_amr = smatch_amr.AMR.parse_AMR_line(amr)
                    parsed_amr = smatch_amr.AMR.parse_AMR_line(parsed_amr_str)
                    smatch_f_score = smatch_test_results.compute_and_add(parsed_amr, original_amr)

                except Exception as e:
                    logging.warn("Exception %s for expected amr %s", e, amr)
                    fail_sentences.append(original_sentence)
                    logging.warn("%s\n with actions %s\n", original_sentence, original_actions)
                    traceback.print_exc()
                if loss is not None:
                    dev_loss += loss.scalar_value()
            accuracy = 0
            if total_predictions > 0:
                loss_dev_words = dev_loss / total_predictions
                accuracy = right_predictions / total_predictions
            accuracies.append(accuracy)

            hist, bins = np.histogram(smatch_test_results.smatch_scores, bins=AMRResult.histogram_beans())
            # plt.hist(smatch_test_results.smatch_scores, hist_bins)
            # plt.show()
            test_result = AMRResult(filter_path, epoch, "test", len(test), accuracy, invalid_actions, bins, hist, smatch_test_results.get_result())
            amr_dynet_results.append(train_result)
            amr_dynet_results.append(test_result)
        print("{} since {} max accuracy {} for {} rounds. Train {} Test {}".format(filter_path, np.argmax(accuracies), np.max(accuracies), rounds, len(train), len(test)))

logging.warning("Results")
logging.warning("Histogram beans: %s", AMRResult.histogram_beans())
output = AMRResult.headers() + "\n"
for result in amr_dynet_results:
    output += str(result) + "\n"
logging.warning(output)